Urban heat island in Amman: AI-based modeling of urban morphology and green infrastructure in mitigating thermal stress

Abstract

Urban heat island effects have intensified in semi-arid cities like Amman, Jordan, due to rapid urban expansion and diminishing vegetation cover. This study develops a predictive framework that integrates remote sensing data, geographic information system–derived urban morphology indicators, and artificial intelligence models to assess and forecast urban heat intensity between 2015 and 2024. Satellite-derived land surface temperature, vegetation cover, and built-up density were used alongside morphological variables such as building height, street width, and road orientation. Several machine learning models, including support vector machines, decision trees, random forests, generalized linear models, nonlinear autoregressive networks, and adaptive neuro-fuzzy inference systems, were tested for predictive accuracy. The adaptive neuro-fuzzy inference system outperformed others with a coefficient of determination of 0.908 and a root mean square error of 0.390. Spatial analysis showed a 12.2% increase in built-up areas and a 9.1% reduction in vegetated land, leading to a significant rise in surface temperatures, particularly in Eastern and Central Amman. The study introduces a novel, high-resolution, machine learning approach for forecasting thermal risks in data-scarce, arid urban regions. Its findings offer actionable insights for urban planners to implement green infrastructure and land use interventions in heat-vulnerable zones.